The future of AI for health insurance

[MUSIC PLAYING] You're listening to Pod Nosis, the pulse of the health care industry. I'm your host, Editor-in-Chief, Aila Ellison. There's no shortage of hype around artificial intelligence and health care. You've heard it all before. AI is going to fix inefficiencies, reduce administrative headaches, maybe even transform patient care. But behind every promise are 100 questions about how it actually works and what it means for the people whose data makes it all possible. This week, senior writer, Paige Mindmeyer, sits down with son Deep Datlani, now CEO of Optum Insight, who at the time of their conversation was chief technology officer at United Health Group. They talk about how one of health care's biggest players is deploying AI right now and the guardrails that have to exist to keep patient data safe. Let's get into it. [MUSIC PLAYING] Sandy, thanks so much for joining us. Paige, thanks for having me. I set the stage for the conversation. I want to get a 30,000 foot view of the state of AI and the pair's face. What's your perspective on this? And how do you see the industry adapting to the rapid change and the conversation around this tech? I think it's a tale of two cities, to be honest. First things, AI is the most powerful transformational force of our lifetime and a huge opportunity for pairs, for providers, for health care to be better for people, to fix itself. There are early adopters who have grappled with AI responsibly and made use cases work and scale. And they've been pragmatic about it, responsible about it. I would like to believe we are one of those as well. And then there are others where, frankly, we need the entire system to pull along and help them make AI work for them. And I believe United Health Group in particular can play a big role in helping the second category also catch up. So it's a tale of two cities. It's a tough road because I still think we are in the first half of the first innings. It's a fast road because every week the capabilities change. And it's a lot of change management at the same time. But it's the most exciting thing we can be working on right now. You said we're at the top of the first ining, essentially, on this technology. Can you talk a little bit about maybe some of the more promising areas for the use of AI and kind of how the team at United Health Group is thinking about those? Yeah, a couple of years back when we met, we were talking about speed and how we could move 100x faster. I think United Health Group so far has moved rapidly in three broad areas. Actually four, I'd say. One is the consumer experience area, so consumer. And I'll talk to you about a few use cases there. And other one is around care and really how the business of health care gets delivered. The third one is around claims, which is really the operational part of the entire payer provider spectrum. And the fourth one is tech engineering. Tech engineering is another critical part where AI has played a big role. So let me give you some examples. This year, we will have 114 million consumer interactions handled by AI bots for us with a equal, a better consumer experience than would have been handled manually. This year, people who come to our digital properties would have searched for a doctor or searched for something specific. And 76 million of those searches will be AI infused. And an AI infused search just so you get a sense is much more likely to find a better answer in a doctor closer to you, a doctor who's appropriate for you. So on from our call centers, we just see over 300 million calls a year. This year, 15 millions of those calls will be obviated because of self-service using these AI bots. So those are some examples of how consumer interactions are becoming seamless. This happened in the airline industry. It happened in the retail industry. It's now happening at pace and scale in the healthcare industry. The apps, the self-service, the AI bots are going fast. And we are now moving to a stage where we can depend on conversational bots, large language models to help us eventually navigate care. And we have some exciting launches coming right to this year. We'll talk about that. The second category is care. So by the end of this year, about 4,000 plus doctors in optimum health would have ambient transcriptions. And ambient transcription is something that's very-- you've probably seen this technology take off among health providers. And then early pilots where we really look at clinical data, claims data, our X data together are helping us define what the next best step for a patient. Of course, a clinician eventually decides that. But the AI is beginning to at least suggest those items. So I think the care part is getting to be super exciting. On the claims part or more, the administrative operational part, United Health Care today is now getting to more than 93% of claims auto adjudicated at first pass. This is a big deal. This wasn't at this higher number for-- I don't think any pair. But now we're getting mid '90s for a spasit adjudication level. There's a lot of AI involved in that. Just the reduction in manual handling time for all of these administrative processes has a ripple effect for costs in the health care system. So our dream there is eventually real-time adjudication, real-time settlement that wipes out all of the administrative areas. And then obviously, technical software development coding is very close to my heart. I have more than 20,000 engineers who are now vibe coding, which means that they're generating software code using generative AI. And every year, they're 20% more productive than the previous year. And that's exciting, right? Because there's so much of health care legacy tech that has to be modernized. The more productive these engineers get, the faster we can modernize UHG systems. And we can help modernize systems of other pairs and providers, no optum does that for other pairs and providers. So I think by the end of this year, we'll have over 10,000 AI builders, people who can build AI. So if I look at those four buckets and those kind of use cases, I'm very excited about how this exponentially compounds for the next couple of years in our quest to solve the health care. Makes sense? Booly, and that's actually a great segue to what I wanted to ask next, which is, you mentioned this as a jumpstart to the future. How much of the work that we're seeing in health care AI is now is foundational to the point where we, in five or 10 years, will see something that would truly be transformative? I love that question. We actually have a framework inside of United Health Group where we're looking at traditional processes. And we're looking at AI1.0. So think of a situation where a traditional back-end administrative process for revenue billing, cycle payment, et cetera, acclaim is automated. So I just told you, oh, claims adjudication rates have gone up. But then we have a third column called AI10.0. The jump from AI1.0 to 10.0 is, well, this is not about automating the process, but what if we could dramatically reimagine the process using the latest generative AI LLM agent framework? So in that whole claim story, we are reimagining the AI10.0 process with something called retail promise. So imagine a situation where it consumer can schedule an appointment online for a particular procedure or treatment, yet the cost estimate of that treatment up front before the appointment, yet there benefits and eligibility check before the appointment, automatically check in as they walk into the clinic, have the provider see them. And as the provider sees them, the provider transcription is automatically coding a claim. And that claim is being settled or promised right away to the provider, even as that clinical appointment ends. This is like a Disney fast pass for the patient and also for the provider. It's a seamless experience. The underpinning, the foundation of that is all of this automation at the back end for the humble claim. And I feel AI1.0, fine, we keep automating all the manual work on the claim. But AI10.0, and we are beginning to launch that early 26. So we're not early 26 is age. It's just a few months away. So we are now getting to a point where we are transforming to AI10.0. Any other experience, look at the provider ambient transcribing experience. Right now it's all right. Dr. Page, you're seeing Sunday, as we talk about my treatment, this is getting transcribed. I have 4,000 plus doctors on it. I'll have 10,000 doctors on it next year. Fine. But next year, what if this same transcription is also beginning to based on our data, nudge the provider, suggest to the provider, the right sort of next best step, and really enable automatic back end processes based on it. That's reimagining the entire provider experience, helps the provider really focus hard on what they're supposed to do. So in every process, and finally consumer, and I love this. We are launching. Next month, ASKI, which is a generative AI chatbot for United Healthcare, UXC app is number one or number two in the App Store, Google, in the medical category. ASKI will first answer, take care of five or six tasks saying, okay, what's my memory, my member ID card, my benefits eligibility, etc, etc. But as the number of tasks increase, ASKI also helps you refueled the restrictions, helps you get your labs. Now it becomes an optimum companion on the optimum health side where you can schedule appointments, you can look at your HSA balances and so on and so forth. The tasks increase. We are also allowing this to be an optimum companion for other healthcare pairs and providers. So as we link up other healthcare pairs and providers, imagine in a couple of years, Paige has a personal companion that knows Paige and it doesn't represent United Health Group or it doesn't represent some other pair of providers. It is Paige's companion, you have your personal health records on it and it just helps you navigate one appointment with one provider to another lab. It just helps you navigate this maze of healthcare that we have created. So that's the vision. And the best part about these vision statements have just stated is they're one year away. They're six months away. They're two years away. They're not talking about a five year vision or a ten year vision because you and I have seen plenty of those in healthcare. And so I am excited about this AI 1.0 to this AI 10.0 jump that we are about to make, which is becoming more and more real every month. Does that make sense? Yeah. Yeah. As you're thinking about this innovation or patients and providers, are they excited about the potential of this technology or they concerned about it and how can organizations like United Health Group help ease them into this arena if they're not quite there yet? Look, I think trust is a huge part of any new technology. So for example, providers, as they adopt ambient transcriptions, our physicians and clinicians have been actually giving us fantastic reviews and feedback. We just launched last week, autosamorization of clinical notes for all the backend clinical nurses that review all exception cases. And I've never seen a faster rollout like 1700 nurses just adopted it and send us such good feedback page. So this whole ambient transcription or summarization of deep clinical notes like that stuff, we're getting fantastic adoption and so on. For many of the other use cases, people demand evidence that this has been responsibly developed. And we are very serious about the responsibility. So we have a responsible use of AI board. So all use cases, we have about a thousand plus use cases in United Health Group. All use cases are under the purview of this board and many of the use cases are systematically reviewed by this board. And the board has clinicians, technologists, ethicists. I don't know. Ethicists was a profession until now. And it is ethicists, a formal ethicist. And so you have a cross functional board, including external advisors on the board that review all these use cases and make sure that there's no bias. There's fairness. There's human in the loop where needed that helps us create and monitor our responsible AI. All AI is developed in our central framework called United AI Studio. And that's very important for us because then I have the right secure data pipes to it, the right secure framework, the right LLM models on it, sexual. This allows us to now deploy AI at scale responsibly and start building trust. Ultimately, once you cross the trust chasm, you have to deliver outcomes. So either I'm helping our operations people or physicians reduce their administrator burden or I'm improving a consumer a provider experience. Something's got to prove or somebody's health is improving or I'm keeping somebody healthy or I'm helping the system. I'll give you an example. Consumers who are searching for doctors are able to complete their tasks about 10% more successfully than they used to before AI was infused in it. That should make a lot of consumers happy. So that's how we're measuring it. Any new technology is tough, but if done responsibly, another rule for responsibly, I bought we never used AI to deny any care. AI denies nothing. Peter. And so those kind of rules help us move forward in this journey. And we are being super careful about it along with our partners. You mentioned kind of a centralized hub for development with this technology, which is given the breadth of the tools available at United Health Group from the UHC side or at Optum. What does that kind of create in terms of a test kitchen for the developers who are looking at ways to deploy this technology? The great question. So technology is a central group at United Health Group. All engineering is done by one group. And so at some level, it's easier to create a single playground for all developers. Now, developers are innovative. People I'm very proud of all our talented developers. We've got to give them the right tool set. So we have all the top large language models, all the top agentic models available in that playground. So you can think of the top five companies with the best models. They're all in United Health Group, but in that safe playground. That allows developers to come back in a democratic way. And we know like, okay, which is the best software development, wipe coding tool. I already know I have telemetry on, hey, out of the 23,000 engineers who touched wipe coding tools in the last one month. Here is the usage. It's clear that one or two tools, one and over here is the productivity increase. Here's the quality increase. Here's the throughput increase. Having that telemetry and observability across all AI models is fantastic. And like I won't publicly rank these models. Hair page. I know you would love me to, but what I can tell you is every three months, the rankings change. What's stuff about this world is, is moving so fast that it's tough to pin down one technology or one LLM model and scale the heck out of it. That's a challenge actually. That's a challenge for every technology leader in health care. You mentioned earlier as well that there's a laundry list of different applications of AI that, you know, health group is currently using it across the enterprise. Are there any projects that were worked on that just didn't pan out and that taught the team significant lessons around kind of development in this area? Oh yeah. No, listen, the path to AI success is littered with a lot of trial and error and failure. And this is not for the faint-tarded. Voice is a good example. Or even these conversational bots are a good example. You see a lot of them already prevalent in retail. For large, let's say retail chains, I want to buy shampoo. I can go on the conversational bot and ask them about. I can't put them out in health care. At let's say 95% accuracy because what are you saying? Are you saying 5% is of the time you're going to be wrong about your benefits and you're going to take health care decisions on that basis. That's a tough thing to do. Health care and I'm so proud of both our engineers and frankly all other health care pairs and providers. They've been very responsible about this. There's startups galore who will try and sell their okay, I have my voice bots do all the calls for you. I'll have these chat bots handle all you and queries. Yeah, but like when you dig deeper into their accuracy, their reliability, the resilience at scale, especially when hundreds of millions of consumers are calling, there's a higher level of accuracy and we've found it very tough. We've stopped this project bit way like this. As I conversational bot getting released next month, that is supposed to be released three quarters ago, but we want to be responsible. We want to be safe. So we tried, we made errors, we tried, we tuned until we were sure that okay, this thing will work. I'm not here to be a perfectionist, but we have to be very pragmatic in health care and responsible in health care. And I think the world of AI, but it's not perfect and that's something we have to compensate for. Makes sense. Yeah, and we've touched a couple times now on just how quickly this technology and its applications are evolving. I want to close where we started and ask you just to talk about what the future looks like if the health care industry is able to really realize the full potential of what this tech can do. I want consumers to have a relationship with their personal conversational chat bot who remembers them who's not Doreen Nemoke used to forget all the time, they could remember them all the time and help them navigate the system so that as a consumer, I don't have to worry about the entire health care system. I just talk to my personal bot. I want providers to focus on the business of the care part of health care. We've created hundreds of other things for them to do, which is wrong. They always dreamed of being the best at care. Let's help them operate at the top of their license. I want administrative here to disappear. There is hundreds of billions of dollars of health care costs all in administrative here and AI is perfect for solving for them. I want technology engineers to be using AI to create breathtakingly new beautiful experiences, beautiful experience. Like, Hell's Case never been associated with the word "beautiful experience". I want engineers to create beautiful experience healthcare. That's the vision. What I am trying to do, especially with Opta Mei, is create a engineer workforce that is deep rooted in solving healthcare problems that can help United Health Group with this journey, but frankly help all pairs and providers in the system because we're not gonna win by being first in this. Everybody will have to get there together as part of a mission statement. Help people live healthier lives and help the system work better for everyone. So my invitation to all my peers, to other pairs and providers is, we may be ahead on a couple of things, you may be ahead of a couple of things. We're not gonna win this race. Let's help each other. Let's have our engineers work towards a common goal and bring that healthcare dream to life in a pragmatic and responsible way. I'm Sandy, I really appreciate the time and your willingness to share your insights with us. (upbeat music) - Thank you for listening to Pod Nosis. I'm Aila Ellison. You can find out more about this topic in our show notes at fearshealthcare.com. Look for podcasts. And don't forget to tune in every Wednesday morning to Pod Nosis, where healthcare is our beat. (upbeat music)